Brian Feldman <gr...@unixhelp.org> writes: > On 13 Jun 1999, Dag-Erling Smorgrav wrote: > > connect() normally uses the 1024-5000 range. Try the following: > > > > # sysctl -w net.inet.ip.portrange.last=40000 > > > > and see if it solves the EAGAIN problem. > > Actually, this is the perfect explanation. I think that this should go in > the FAQ. Why in the world are we limiting the ports we can use other than > from 1-1023?
I have no idea. The only use I've ever had for this is the predictability of FTP data port numbers, since ftpd will normally use the high range (49152-65535). Useful for designing firewalls, and if you don't like opening such a wide port range, you can whittle it down to a handful (say, one thousand) of ports with judicious use of sysctl. I couldn't find any reference to this in RFC 793, 1122, 1123, 1700 or 2577. Remind me to build a glimpse index of my RFC collection... This still doesn't explain select()'s failure to time out, although I should point out that you forgot to increase FD_SETSIZE before including <sys/types.h>, and I'm not sure your computation of highestsock is correct, since there's no guarantee that fdvec is sorted. I haven't tested your code though, I just looked over it quickly. BTW, you should check for errno == EINTR when select() returns -1. DES -- Dag-Erling Smorgrav - d...@flood.ping.uio.no To Unsubscribe: send mail to majord...@freebsd.org with "unsubscribe freebsd-hackers" in the body of the message
